PREDICTING CHORDAL TRANSITIONS IN JAZZ: The Good, the Bad, and the Ugly

Instance-based machine learning techniques have been applied to predict the next chord in a piece of jazz music. This study compares off-line (trained on a static data base of songs) and on-line (trained on-the-fly) prediction techniques. A hybrid model combines both on-line and off-line models. On average, chords in a jazz piece are predictable 53% 17% of the time using the hybrid model. Successful predictions were defined as exact matches of chord root and type.